Field notes: the week of October 16, 2026

A client asked us this week why their lead volume dropped 30 percent while their ad spend held flat. The answer turned out to have nothing to do with the ads. Their intake form had been silently rejecting submissions from one browser version for eleven days. Nobody noticed because the dashboard they watched counts clicks and cost, not completed forms. This is the kind of thing that gets lost when you measure the top of the funnel and assume the bottom takes care of itself.

Here are the notes from the week.

The gap between "working" and "verified"

The form bug is worth sitting with. Everyone involved was competent. The ads ran. The landing page loaded. The CRM showed new records, just fewer of them. Each person checked their own layer and found it fine. No single owner checked the whole path from click to qualified lead.

We now run a weekly end-to-end test on every lead path we manage. Someone fills out the actual form, on an actual phone, and watches the submission land in the actual CRM with the correct source tag. It takes about ten minutes per client and it catches things that no analytics view will surface, because analytics can only report on events that fire. A form that fails silently fires nothing.

Submit a real test lead through the live form, not a staging copy.
Confirm it arrives in the CRM with the right source attribution.
Confirm any automated follow-up (email, text) actually sends.
Check it on at least one older browser and one phone.
Log the date so you can see when something was last confirmed working.

The lesson is not "test your forms." Everyone says that. The lesson is that "it was working last month" is not evidence that it works now. Software you did not change can still break, because the browsers, plugins, and services around it change underneath you.

Attribution honesty with local service businesses

We onboarded two local service businesses this week, and both came in with the same belief: that every call and form could be traced cleanly back to a channel. They had been sold reporting that implied this. It is not true, and pretending it is true leads to bad budget decisions.

A homeowner sees a Google ad on Monday, forgets about it, searches the business name directly on Thursday, and calls from the second visit. Standard attribution credits the direct visit and gives the ad nothing. Multiply that across dozens of interactions and you get a report that systematically undervalues the channels that create demand and overvalues the ones that capture it.

Attribution is a model, not a measurement. The job is to pick a model honest enough to make better decisions, not one precise enough to feel certain.

For local businesses we lean on a few less glamorous signals that are harder to fake: call tracking numbers per channel, a required "how did you hear about us" field that a human reviews, and a simple before-and-after on total inbound volume when we turn a channel on or off. None of these are perfect. Together they give a picture that holds up better than a dashboard claiming four-decimal precision.

What the grader keeps finding

We push a lot of prospects through the free site grader before any conversation, and a pattern repeats. The things that hold sites back are rarely exotic. This week's batch surfaced the usual suspects:

Page titles that describe the company instead of what a visitor searches for.
No clear primary action above the fold, or three competing ones.
Phone numbers that are not clickable on mobile.
Slow largest-contentful-paint caused by a single oversized hero image.
Contact forms with more fields than the business actually needs.

Each of these is a small fix. The reason they persist is that nobody owns the whole experience. A developer built the site, a designer picked the hero image, a marketer added the form fields, and no one is responsible for whether the combined result converts. That ownership gap is more expensive than any one mistake.

A note on form fields

The field count point deserves emphasis because it is the easiest win we see. Every field you add filters out some people who would have converted. Sometimes that filtering is worth it, a qualified lead is better than ten unqualified ones. But most businesses add fields out of habit, not strategy. If your sales team can qualify on the first call, you probably do not need to qualify on the form. Ask for the minimum that lets someone follow up, then earn the rest of the information in conversation.

We spent part of this week on our own MCP servers, the pieces that let Claude work directly inside our project manager and portal. The practical effect is that a partner can ask for the status of every client's lead paths and get an answer assembled from the live data, rather than someone manually checking each account.

There is a reason we build this for ourselves first. We do not recommend tooling to clients that we have not run against our own operation under real pressure. A lot of marketing advice fails because it was never load-tested by the person giving it. When something we build survives contact with our own messy, multi-client reality, we trust it enough to put in front of a client. When it does not, we find out on our own time instead of theirs.

The weekly end-to-end lead test came out of exactly this. It started as a chore we did for ourselves after getting burned, and it became a standard because it kept paying off. The best processes we have were almost all born from a specific failure we did not want to repeat.

The through-line

Three of the things above are the same thing wearing different clothes. The silent form bug, the attribution fantasy, and the unowned site experience all come from assuming that something is fine because no one is actively reporting that it is broken. Absence of alarm is not the same as health. The teams that win treat verification as a recurring job, not a one-time setup, and they stay honest about the limits of what they can measure.

If you want a fast, outside read on where your site leaks before you talk to anyone, run it through the free grader. If you would rather have a partner whose pay depends on fixing what the grader finds, apply for partnership and we will take a look at the whole path, not just the top of it.